Continuous gravity recording with Scintrex CG-3M meters: a promising tool for monitoring active zones

Transcription

1 Continuous gravity recording with Scintrex CG-3M meters: a promising tool for monitoring active zones Sylvain Bonvalot, Michel Diament, Germinal Gabalda To cite this version: Sylvain Bonvalot, Michel Diament, Germinal Gabalda. Continuous gravity recording with Scintrex CG-3M meters: a promising tool for monitoring active zones. Geophysical Journal International, Oxford University Press (OUP), 1998, 135 (2), pp < /j X x>. <insu > HAL Id: insu Submitted on 17 Aug 2016 HAL is a multi-disciplinary open access archive for the deposit and dissemination of scientific research documents, whether they are published or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers. L archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.

2 Geophys. J. Int. (1998) 135, Continuous gravity recording with Scintrex CG-3M meters: a promising tool for monitoring active zones Sylvain Bonvalot,1,2 Michel Diament2 and Germinal Gabalda1 1 ORST OM, L aboratoire de Géophysique, 32 Avenue Varagnat, Bondy Cedex, France. bonvalot@bondy.orstom.fr 2 Institut de Physique du Globe, L aboratoire de Gravimétrie et Géodynamique, 4 Pl. Jussieu, Paris, France Accepted 1998 June 16. Received 1998 June 12; in original form 1997 November 10 SUMMARY We acquired continuous series of microgravity measurements using several Scintrex CG-3M gravity meters for several weeks in The meters with 1 mgal resolution were installed side by side in a stable reference station at the ORSTOM research centre to perform identical data acquisition. We present and compare the instrumental responses obtained for the various gravity meters (measurement series of gravity field, standard deviation, internal temperature, tilts) and analyse their correlation with simultaneous recordings of meteorological parameters. The data have been processed in order to (1) establish the mid- to long-term relative stability and the accuracy of the instruments, (2) estimate the contribution of instrumental effects to gravity data measurements and (3) quantify the amplitude of the time variations of the gravity field that might be detected with such instruments. This study emphasizes the sensitivity of some instrumental responses of the Scintrex CG-3M gravity meters (such as internal temperature or tilt) to local atmosphericpressure variations. This sensitivity can lead to non-negligible perturbations of the gravity measurements through automatic corrections applied in real-time mode by the integrated software. We show that most of these instrumental artefacts can be easily removed in data post-processing by using simultaneous atmospheric-pressure data. After removal of an accurate Earth tide model, the instrumental drift and the instrumental effects, the temporal series are compared by computing differential signals. These residual signals obtained over a period of several weeks exhibit the following characteristics: ( 1) the gravity residuals have a maximum amplitude ranging from 5 to 10 mgal and from 10 to 15 mgal for filtered and unfiltered data, respectively; and (2) the standard error, tilts and internal temperature measurements of the various gravity meters are very consistent; their respective residual amplitudes are ±2 mgal, ±3 arcsec and ±0.05 mk. In order to calibrate the gravity meters precisely in the measurement range used in this study, we have measured a calibration line established in the framework of the fourth intercomparison of absolute and relative gravity meters. This calibration was achieved with an accuracy of 5 mgal. This result is consistent with other field tests already performed with such gravity meters. In addition, we also checked the accuracy of the tilt sensors by increasing the electronic read-out by a factor of 10. The tilt response of the whole gravity meter to a small induced inclinometric variation indicates that the precision of the tilt measurements is about a few tenths of an arc second. This study reveals that temporal variations of the gravity field could potentially be detected in the field with an accuracy of about 5 15 mgal by permanent networks of Scintrex CG-3M gravity meters set up a few kilometres apart. This result is of particular interest in field surveys of temporal gravity changes related to some environmental or geodynamical processes, where the expected gravity variations are greater than a few tens of mgal. In particular, in volcanological applications, the continuous monitoring of active volcanoes with such permanent networks of gravity meters co-located with subcentimetre-accuracy GPS receivers should be very helpful to understand internal RAS

3 Continuous gravity recording 471 magmatic processes better and to detect possible gravity and inclinometric signals occurring during pre-eruptive phases. In this field, continuous microgravity recordings associated with classical reiteration networks will probably improve hazard mitigation in the near future. Key words: gravity meter, microgravity, Scintrex CG-3, volcano monitoring. 1 INTRODUCTION amplitude of the residual gravity variations ranges from several tens to several hundreds of mgal. The observed variations are The study of the time variations of the Earth s gravity field usually episodic and occur shortly before or after phases of can provide fundamental information on the internal dynamics volcanic activity. The study of these variations for which of the globe at different scales and also on some meteorological microgravity methods are more and more often used, is of phenomena (Lambeck 1981; Goodkind 1986; Torge 1981; particular interest for the monitoring of volcanoes. In addition, Hinderer, Legros & Crossley 1991; Groten & Becker 1995). a volcanic eruption precursor signal was clearly pointed out Depending on their origin, these variations occur over long or for the first time on the Poas volcano (Costa Rica) in 1989 short periods of time, periodically or temporarily. They are (Rymer & Brown 1989). detected by repeated relative or absolute measurements made Temporal gravity variations on volcanoes are usually at reference stations for the long-term phenomena, and by detected by repeated measurements. Microgravimetric and continuous measurements for the short-term phenomena. geodetic networks, with stations located in stable zones as well These studies rely on highly accurate observations that currently as in active zones, are used for these measurements. The time can be performed under laboratory-like conditions with interval at which the networks are reoccupied, typically from the use of absolute gravity meters or superconducting relative several weeks to several years, allows the recording and the gravimeters, whose resolution can reach mgal (1 mgal= study of long-period variations which are related to long-term 10 nm s 2). Such accuracy cannot be obtained in the field magmatic phenomena. In order to detect shorter-term varibecause of the instrumental limitation on one hand (field ations, either the networks should be occupied more often or instruments are more robust, more portable but less accurate), continuous recordings of the gravity field should be made. A and because of a greater exposure to external perturbations of few time-series have been acquired on several active volcanoes the measurement sites (meteorological effects, microseismic in this way: for example, on Etna, Italy (Berrino et al. 1995; activity, etc.) on the other hand. However, temporal gravity De Meyer, Ducarme & Elwahabi 1995; Budetta, Carbone & changes of relatively large amplitude (greater than a few tens Rymer, personal communication) and on Merapi, Indonesia of mgal) can be studied under local conditions on tectonically (Jousset et al. in press). Some of these time-series showed a active or volcanic zones. Recent instrument-technological correlation between the gravity field variations and the mag- developments in microgravimetry and in related domains such matic or seismic activity. Most of the time, these data have as GPS geodesy now allow us to acquire more easily a larger been acquired at a single site with only one instrument. amount and higher quality of field data ( higher sensor accu- Therefore, it is difficult to connect the data with certitude to racy, numerical data acquisition and processing, higher storage the various events, because of limiting factors such as the small capabilities, etc.). Nowadays, these improvements allow us to numbers of time-series, too short a period of recording, and detect and to study phenomena of smaller amplitudes or the importance of instrumental effects. In order to avoid some occurring over shorter periods, for which the resolution limit of these shortcomings, the optimal method would consist in of the instruments can be reached. A good knowledge of the deploying permanent networks of microgravimeters distributed instrument responses and of the various factors that can over the active zone and over a stable zone used as a reference. influence the data quality is then indispensable in order This method has several advantages for volcanic monitoring, to evaluate the actual accuracy of the measurements and, including: therefore, the order of magnitude of the observable phenomena. (1) a more accurate analysis of the time variations that As an example, temporal gravity studies in tectonically active could lead to a possible detection of eruption precursors; or volcanic zones require at the same time very accurate (2) a continuous recording of the data even during the instruments and well-defined data acquisition and processing active periods, limiting the risks for the operators; procedures, in order to minimize the various error sources and (3) a diminution of the number of in-field tasks (network to reach the required accuracy. Regarding the microgravity reoccupations) and automation of the monitoring tasks. monitoring of active volcanoes, these methodological aspects as well as numerous examples of applications have been So far, in-field microgravity surveys and the abovementioned discussed in publications presenting the state of the art in this type of continuous series on volcanoes have mainly domain (Rymer & Brown 1986; Tilling 1989; Eggers 1987; been realized with relative gravimeters of the LaCoste & Berrino et al. 1992; Rymer 1994, 1995). These temporal gravity Romberg type, whose resolution varies from 10 to 5 mgal for variations are related to changes in the internal structure of the G and D models, respectively (LaCoste & Romberg 1991). the systems (mass redistribution, density changes) or to ground Many studies have confirmed the accuracy and the stability of motions (altitude change of the measurement site) in response to a magmatic activity. After removal of the component due to a possible deformation of the topographic surface, the these instruments for the detection of gravity field variations in active zones. The accuracy obtained by reoccupying the networks with such instruments usually varies from 15 to

4 472 S. Bonvalot, M. Diament and G. Gabalda 20 mgal (Rymer 1989, 1994; Torge 1989). These instruments reoccupation on several active volcanoes: Masaya (Bonvalot can also be equipped with an automatic system of measurement et al. 1995); Piton de la Fournaise (Bonvalot et al. 1996; which allows the continuous recording of analog or digital Diament et al. 1997), Merapi (Diament et al. 1995; Jousset data. A model with a limited range of measurement of 12 or 1996), La Soufrière (Diament et al. 1997); Etna (Budetta & 14 mgal (ET model) was especially created for Earth tide Carbone 1997), for instance. studies (LaCoste & Romberg 1996). These different models of At present, there is no published study of continuous gravimeter (D, G and ET types) could also be modified by recordings of the time variations due to geodynamic effects integrating an electronic feedback system with the original (seismic or volcanic active zones) using the Scintrex CG-3/3M sensor (Harrisson & Sato 1984; Van Ruymbeke 1985; Vaillant recording system. Nevertheless, the instrument is particularly 1986). Under good conditions, the accuracy of the Earth tide well suited to this field of application. Therefore, a comparative recordings made on these instruments is estimated to be 0.5 study has been carried out on several instruments with microgal or 1 mgal, on the basis of the standard deviations defined for sensitivity in order to check the middle- and long-term behav- the hourly values (Torge 1989, p. 378). These observations iour and stability of the Scintrex CG-3M gravimeter responses. confirm that the technical characteristics of the LaCoste & Continuous recordings have been acquired on the same site in Romberg instruments are well suited to the continuous moni- order to analyse the influence of the instrumental effects on toring of volcanoes, although these instruments were developed each device and to evaluate the expected accuracy for this type a few decades ago. of study. A new generation of relative gravimeters was conceived by Scintrex Ltd at the end of the 1980s. The AutoGrav CG-3 and 2 THE SCINTREX CG-3/3M GRAVITY CG-3M meters are based on the use of microprocessors, which METER allowed the automation of the measurements and their processing (Hugill 1990). These instruments, with resolutions of 5 mgal and 1 mgal, respectively, can be used in two different 2.1 Functional principle modes: an in-field mode allowing the acquisition of discrete The measurement of the gravity field in this instrument is based measurements, and a cycling mode for continuous data on a capacitive measurement of the extension of a vertical recording. Their technical characteristics make these instru- quartz spring. This geodetic-type device allows a worldwide ments useful for various applications of relative gravimetry measurement of the gravity field over a range of 7000 mgal based on the study of the spatial and temporal variations of without resetting. Currently, its resolution reaches 5 mgal for the gravity field. Therefore, they can also be used in microgravity the standard version (CG-3 model) and 1 mgal for the microgal studies in volcanology, for discrete measurements as well version (CG-3M model). At a given station, the gravity field as for continuous recordings. Several recent studies have relative value is determined by a series of measurements confirmed the potentialities of these instruments for the micro- (generally single measurements) performed at a sampling gravimetric survey of superficial structures and for network rate of 1 Hz. The mean value and its standard deviation are Figure 1. Simplified acquisition scheme for the Scintrex CG-3/3M gravimeter (modified from Scintrex 1995). ET C is the software-computed Earth tide correction based on the Longman (1959) algorithm. The parameters TILT XS and TILT YS are the tilt sensor sensitivities adjusted by the operator (see text). T EMPCO is the temperature correction factor determined by the manufacturer (see text). GCAL 1 and GCAL 2 are the first- and second-order calibration factors of the gravity signal and DRIFT is the correction factor used for the instrumental linear drift correction.

5 computed from the single measurements after rejecting outliers. In addition, this instrument is equipped with tilt and internal temperature sensors, providing real-time numerical corrections of the gravity measurements. These corrections are applied for a range of internal temperature of ±2 mk and for a range of tilt of ±200 arcsec. Then, the numerical data are stored internally and can be transferred to a computer through an RS232port.Fig.1showsadiagramofthedataacquisitionand processing system. Technical details of the instrument and the acquisition procedures can be found in Hugill (1990), Scintrex (1995) and Siegel, Brcic & Mistry (1993). The Scintrex CG-3/3M gravimeter can be operated either in field mode or in cycling mode. Depending on the operating mode, data acquisition is triggered either by an operator, once the meter has been installed temporarily, or automatically at a pre-defined sample rate (typically starting from 1 point min 1), for a fixed device. These modes allow for measurements of both spatial and temporal variations of the gravity field. 2.2 Main domains of application Various methodological aspects of time variation studies using continuous recording have been tackled by Goodkind (1986). Time-series acquired from superconducting gravimeters were analysed to point out and to interpret the residual variations observed after Earth tide and instrumental-drift corrections had been applied. Such residual variations can be related to geophysical, instrumental or atmospheric factors. In order to determine the precise cause of these variations, it is first necessary to look for a possible correlation between the gravity observations and other parameters simultaneously recorded. In particular, the analysis of the correlation should allow one to discriminate between the variations related to external factors (meteorological, geodynamic) and the variations due to instrumental effects (internal temperature, tilt). In consequence, the identification of instrumental artefacts is a fundamen- tal step in the analysis of gravity field continuous recordings. Indeed, the amplitudes of these instrumental effects are comparable to or greater than those of the actual signals. Goodkind (1986) has also shown the advantage of using several instruments of the same type in order to make simultaneous recordings of the gravity field. This approach is the only rigorous way to determine the instrumental noise limits and the instrumental drift. By running at least two meters at the same site, their relative stability can be assessed and controlled periodically. Then, the meters can be operated at remote sites. Differential recordings can allow the detection of time variations with a previously determined accuracy. Another benefit of this method is to eliminate some of the local gravity variations that are not correlated with instrumental effects. For instance, this is the case for uncorrected Earth tide periodic residuals or for local variations of the atmospheric pressure. Both can be minimized by the computation of differential The main purpose of this instrument is for measurements in geophysical prospecting (gravimetric and microgravimetric measurements in oil and mining prospecting, civil engineering, etc.). Several real-time acquisition and processing procedures have been included by the manufacturer to increase the speed and efficiency of the data collection, and to ease the postprocessing of the data (theoretical correction for the lunar solar tide applied for a given latitude, removal of the long-term instrumental drift determined by continuous recording, etc.). This meter can also be used for continuous measurements of the Earth s gravity field. Ducarme & Somerhausen (1997) recently analysed the Earth tide recorded in Brussels by a CG-3M gravimeter over an eight-month period of time. Despite a strong instrumental drift (a few tenths of a mgal day 1) and a relatively low resolution for this type of study, the results, as well as a comparison with other gravimeters (LaCoste & Romberg, GWR superconducting), confirm that this instrument is also suitable for Earth tide studies. Other applications involving the measurement of time variations observed at a local scale, such as in active seismic or volcanic zones, can be also envisaged. The performances of the Scintrex CG-3/3M gravimeters have been compared to those of other relative gravimeters usually used for such purposes (LaCoste & Romberg type). Examples include the fourth intercalibration Continuous gravity recording 473 of absolute and relative gravimeters (Jousset et al. 1995) and field measurement results (Budetta & Carbone 1997). These studies have shown that a repeatability of the measurement of the order of 5 10 mgal could be obtained with these instruments on network measurements. Such results have also been recently confirmed by another comparison of LaCoste & Romberg and Scintrex meters (Kauffmann & Doll, in press). 3 IMPLEMENTATION OF CONTINUOUS RECORDING 3.1 Methodological aspects and study objectives Figure 2. Difference between two theoretical models of the Earth tide between 01/16/97 and 02/20/97 at ORSTOM Research Centre in Bondy ( N, E). The residual signal is the difference between the correction computed by the Scintrex CG3 software [Longman (1959) algorithm] and the correction computed using the amplitude and phase coefficients determined by the Royal Observatory of Belgium (MT80 software).

6 474 S. Bonvalot, M. Diament and G. Gabalda Figure 3. Raw gravity recordings acquired on several Scintrex CG-3M gravimeters side by side (Bondy, from 01/16/97 to 02/20/97) with no internal drift correction applied. The data acquired at a time interval of 2 min, were undersampled at a sampling rate of 1 point hr 1. The drift and offset parameters were set to zero in order to quantify their actual instrumental drift. Table 1. Instrumental parameters for three Scintrex CG-3M gravity meters deduced from continous gravity recordings at Bondy from 1997 January 16 to 1997 February 20. (a) Computed longterm intrumental drift values using the linear and quadratic models. ( b) Averaged admittance values between internal temperature and gravity computed from long-period recordings. (c) Admittance values between tilt responses and atmospheric pressure variations within the studied area. Instrument serial number (a) Instrumental drift Linear model Correlation coefficient M (mgal) M (mgal day 1) Quadratic model Correlation coefficient M (mgal) M (mgal day 1) M (mgal day 2) (b) Averaged long-term admittance between temperature and gravity (mgal mk 1) (c) Admittance between tilt X and atmospheric pressure (arcsec hpa 1) Admittance between tilt Y and atmospheric pressure (arcsec hpa 1)

7 Continuous gravity recording 475 signals. As a matter of fact, Goodkind (1986) showed that the acquired on several instruments set in similar recording conditions influence of the atmospheric pressure was less than 0.1 mgal at the same site have been compared. Four Scintrex for two meters 10 km apart. This result was confirmed by CG-3M gravimeters (numbered , , , Merriam (1992), who argued that 90 per cent of the gravimetric and ) with a sensitivity of 1 mgal, bought by ORSTOM effects originating from the atmosphere are constant over a and the Institut de Physique du Globe, Paris, between km radius area. However, in uneven regions where atmospheric and 1996, were used in this experiment. The main objectives effects may occur more frequently, a smaller radius of this study were should be considered. The various residual contributions due to the Earth tide or to atmospheric effects can potentially be (1) to establish the relative stability and the accuracy of eliminated by taking a difference between signals recorded by the meters; two gravimeters a few kilometres apart. (2) to estimate the contribution of the instrumental effects The superconducting gravimeters used by Goodkind (1986) for this type of instrument; had a better resolution than field gravimeters or microgravime- (3) to quantify the amplitude of the variations that could ters. This is due to a quasi-null instrumental drift and to a be detected by a gravimetric differential method. very low noise in the measurements obtained on cryogenic gravimeters, which can detect very-small-amplitude time variations (sub-mgal level) (Hinderer, Crossley & Xu 1994). 3.2 Experimental set-up However, for some variations of geodynamic origin, such as For several weeks in 1996 and 1997, three instruments set for those related to volcanic activity, the expected amplitudes do continuous recording were installed side by side in a vault at not necessarily require the use of cryogenic gravimeters. the ORSTOM research centre in Bondy ( N, Besides, these gravimeters are poorly adapted to difficult field E). The fourth meter, not available for a long period conditions such as those encountered on volcanoes. In order to assess the Scintrex CG-3/3M gravimeters capacities under continuous recording conditions, time-series of continuous recording at that time, was used for calibration purposes. Recordings were made at a sampling rate of one point per 2 min (cycle time 120 s, read time 90 s, calibration Figure 4. Residual gravity signals obtained after removal of a linear drift and of an accurate Earth tide model. It can be seen that the linear model does not fit the instrumental drift over this period of time (see text and numerical values in Table 1).

8 476 S. Bonvalot, M. Diament and G. Gabalda frequency 1/12 sample). First, tuning of the sensitivity and Observatory (Ducarme, personal communication). For each temperature, and tilt corrections were performed on all instruments, wave group, the amplitude and phase coefficients of this model according to the recommendations made by Scintrex were determined on the basis of the analysis of the tide observed (1995). This tuning ensures that the measurements are properly at the Bureau International des Poids et Mesures in Sèvres, corrected. In order to quantify the actual instrumental drift, France. Because of the closeness of our measurement site (less the drift correction parameters and the offset were initialized than 20 km away) to that reference, the same coefficients were to zero. To ease the reading of the time-series recording over applied for the computation of the tide correction. Fig. 2 shows a long period of time (greater than a month) and to facilitate the deviation between the correction so determined and the one their comparison with other recordings made simultaneously computed by the Scintrex software for the time period covered (meteorological), the gravity signals were later under-sampled by the gravimetric recordings. Taking into account all effects at a sampling rate of one point per hour. To study the residual related to the tide in this accurate model leads to an improvement responses of the various gravimeters, contributions linked to of up to 10 mgal with respect to the standard model the earth tide and the instrumental drift were first removed applied in the software. This precise Earth tide correction was from the recordings. computed by means of the interactive CG3TOOL software 3.3 Earth tide correction especially designed for Scintrex CG-3/3M data processing (Gabalda & Bonvalot 1997). The Earth tide correction computed by the Scintrex software is based on an algorithm developed by Longman (1959). The 3.4 Instrumental drift correction is applied in real time to the measurements. The Raw recordings for the time period between 1997 January 16 model used in this software was not accurate enough for and 1997 February 20 are presented in Fig. 3. No tide correc- microgravity studies. Thus, a new correction was computed tion has been applied to these data. The instrumental drifts using the theoretical model produced by the Belgium Royal observed during this time period on the various meters are all Figure 5. Residual gravity signals obtained after removal of a quadratic drift and of an accurate Earth tide model. It can be seen that the quadratic model correctly fits the instrumental drift for all instruments (see text and numerical values in Table 1).

9 meter Nevertheless, the amplitude of these variations is large and oscillates between and mgal for devices and and between and mgal for device When compared to other field gravimeter instrumental drifts, that of the CG-3M gravimeters appears quite large. According to Hugill (1990) and Scintrex (1995), this long-term drift is high when the sensors are manufactured but should decrease to a value of 0.2 mgal day 1 after several years of usage. After correction of the measurement series for an average long-term drift, the residual drift should be lower than 0.02 mgal day 1. The relationship between the long-term drift and the instru- ment age has not been verified in this study, where the strongest drift was observed for an instrument purchased in 1991 (Fig. 3). However, it can be assumed that a change in the instrumental drift occurred in 1995 after this instrument had been serviced by the manufacturer. The long-term instrumental drift characteristics of the Scintrex CG-3M gravimeters are discussed later. positive, but slightly different from each other. After the lunar solar tide effect had been corrected on the basis of the previous computation, the instrumental drift of the various meters was quantified using two drift models, the first linear and the second quadratic. In the first computation, a drift trend of the form y= M +M t was removed. The long-term linear drift values 0 1 defined in this way can reach several tenths of a mgal per day. They vary slightly from one instrument to another, as shown in Table 1. The residual gravity variations (Fig. 4) vary between 0.05 and 0.2 mgal depending on the instrument. They clearly demonstrate that the linear model does not fit the long-term instrumental drift. However, the linear model is more suitable to shorter time windows, up to 10 days. This result conforms to the manufacturer s specifications for these instruments. Residual signals obtained by removing a quadratic drift of the form y=m +M t+m t2 from the recordings are shown in Fig. 5. The coefficients for this model are reported in Table 1. The shape and amplitude of the residual signals confirm that the quadratic model correctly fits the instrumental drift over time periods greater than several days. The residual variations are relatively consistent among the different instruments, even though they seem strongly amplified (by a factor of 2) for Continuous gravity recording Instrumental noise and microseismicity As shown in Fig. 5, a high-frequency noise affects all residual gravity signals. Its amplitude is estimated to be about 0.01 mgal for the first set of recordings (days 16 34) and about mgal Figure 6. Standard errors obtained for the three instruments side by side (Bondy, from 01/16/97 to 02/20/97). The similarity of the various responses indicates that all instruments have the same sensitivity to external noise.

10 478 S. Bonvalot, M. Diament and G. Gabalda for the second set of recordings (days 34 52). Fig. 6 shows the instrumental factors. In order to assess the contribution of standard errors computed for the corresponding measurement instrumental factors, a possible correlation between the gravity series. For each instrument, these errors lie within the range residuals and other parameters measured over the same time mgal, and are slightly higher for the second set period was investigated. We considered in this comparison the of recordings. These errors are derived from the standard- parameters recorded by the instrument such as the internal deviation values of each measurement computed by the temperature and the sensor inclination. In addition, we used Scintrex software. Each measurement is defined by an arith- local meteorological data recorded close to the studied area. metic mean over a series of N values and the corresponding Fig. 7 displays atmospheric pressure and temperature vari- standard deviation SD. Individual values with an error more ations observed over the same period of time. These data, than four times larger than the standard deviation are rejected acquired at a sample rate of one point per hour, come from a from the mean computation (an option available in the Scintrex station of the French meteorological array (Roissy station), software), leading to an actual number of values included in located less than 15 km away from the measurement site. We the computation, DUR, that is smaller than N (Scintrex 1995). first examine the thermal and inclinometric instrument On the assumption that the noise is normally distributed, the responses, since the Scintrex software corrects each gravity error in the measurements series is estimated as follows: measurement in real time for the internal temperature and the sensor tilt. Err= SD This error includes both the instrumental accuracy of the 4.1 Thermal responses acquisition system and the surrounding microseismic noise Fig. 8 presents the internal temperature variations recorded acting on the vertical spring at various frequencies. However, for the various gravimeters after removal of a linear trend. The a homogeneity of the measurement errors for all gravimeters parameter varies in a consistent manner for all the meters, but is observed over time. This response demonstrates the identical with slightly different drift values ( mk day 1). sensitivity of the different instruments to the same external Short-wavelength variations, identified on gravimeter , phenomena. Therefore, these instruments can be considered to can also be observed on the other meters, but strongly attenu- be an accurate tool to measure the microseismic activity level ated. A comparison with the meteorological recordings (Fig. 7) at a given site. shows that these thermal variations are not correlated with the local variations of the temperature, but that they are perfectly correlated with atmospheric-pressure variations. This correlation is the result of the direct effect of the external 4 INSTRUMENT RESPONSES Residual variations obtained after Earth tide and instrumentaldrift corrections may be linked to geophysical, atmospheric or pressure on the temperature and reveals either a poor insulation of the gravimeter (the temperature sensor is located close to the gravity sensor in the thermostatically controlled Figure 7. Variations of the temperature and atmospheric pressure observed in the area of Bondy during the period of gravimetry recordings (at the site of Méteo France, Roissy, from 01/16/97 to 02/20/97). The pressure recorded at the station elevation and the pressure reduced to sea level are displayed as solid and dashed lines, respectively.

11 Continuous gravity recording 479 Figure 8. Variations of the internal temperature of the gravimeters after removal of a linear drift. A comparison with simultaneous meteorological recordings shown in Fig. 7 demonstrates a strong correlation between the atmospheric pressure and internal temperature for meter chamber) or an important sensitivity to pressure variation of variations already identified for this instrument (Fig. 5). This some of the acquisition system elements (electronic components). can be explained by the fact that the internal temperature is Such relationship has already been observed during used to correct in real time the measurement series of the previous laboratory studies (Jousset 1996). thermal variations made in the sensor enclosure (Fig. 1). The In order to suppress this effect, the transfer function linking correction factor (TEMPCO), determined experimentally by the local atmospheric pressure to the instrument internal the manufacturer for each instrument, ranges from 0.1 to temperature has been computed for device This function 0.15 mgal mk 1. It is applied in a temperature window was determined by applying a linear regression between both from 2.0 to +2.0 mk. Hence, perturbations of the internal parameters (Fig. 9a). The result of this regression shows that temperature of the sensor can induce large gravimetric variations. a factor of mk hpa 1 could be used to remove the In the case of meter , the temperature correction correlation between the two signals. After applying this correc- was recomputed using the instrument s own correction factor tion, the thermal signal becomes comparable to those obtained TEMPCO ( mgal mk 1), applied to the corrected on the other devices (Fig. 8). It should be noted that this effect thermal signal obtained by using the previously computed of atmospheric pressure can hardly be detected in the thermal coefficient ( mk hpa 1). Fig. 9(c) shows that the or gravimetric recordings from the other instruments. The resulting residual signal becomes comparable to those of higher sensitivity to pressure of meter may be related the other devices (Fig. 5). either to its older age (first generation of CG-3M, with a less robust acquisition system than the later instruments) or to the fact that the device is thermostatically controlled at 55 C 4.2 Inclinometric responses ( high-temperature option, CG-3MH) while the others are The inclinometric responses of the gravimeters are displayed controlled at 45 C (standard option). in Fig. 10, for two perpendicular axes X and Y. These responses The internal temperature variations over a short period of are homogeneous: the long-term drifts, as well as the daily time seem to be well correlated with the residual gravity variations, are identical on all devices. No drift is noticeable

12 480 S. Bonvalot, M. Diament and G. Gabalda Figure 9. Influence of the atmospheric pressure on the internal temperature of the gravimeter (see text and Figs 7 and 8). (a) Regression model between the two parameters. The corrected temperature (b) and gravity (c) signals should be compared to the raw data in Figs 8 and 5, respectively. along the Y axis for the whole recording period, while a weak then produce a corresponding gravity correction. This correction drift of about 20 arc seconds on the X axis is seen for the is applied to gravity readings for tilt variations within a same time period. These observations indicate, on one hand, ±200 arcsec range. The gravity value corrected in this way, the relative stability of the measurement site, and on the other R(h, h ), can be expressed as (Scintrex 1995) x y hand, the excellent stability of the inclinometers used in the R(h, h )=RU(0, 0) gt(cos h cos h cos X cos Y ), (2) gravimeters. At short periods, there is a weak correlation x y x y between the tilt responses and the atmospheric-pressure where RU(0, 0) is the uncorrected gravity reading for h = x recording (Fig. 7). This correlation, occurring without any h =0, gt is the mean gravity value at sea level, h and h are y x y phase lag, might be related to the pressure effect of the air the tilt values of the gravity sensor in the x and y perpendicular column on the external enclosure of the gravity meters. The directions, and X and Y are the corresponding values displayed coefficients derived from a linear regression analysis of this by the software. correlation are shown in Table 1 and Fig. 11. The admittance According to Scintrex (1995), the perfect tilt adjustment values, ranging from 0.09 to 0.2 arcsec hpa 1, lead to a slight condition is the coincidence of instrument zero tilts, as defined correction of the inclination signals, of less than 5 arcsec. by the digital read-out of the bubble level and the tilts referred According to the relation between gravity and tilt variations, to the horizontal as defined by the maximum sensor output. these corrections are lower than the 1 mgal level. In view of For correct operation, the tilt adjustment should be periodically the very low amplitude of this effect of the atmospheric pressure checked following the manufacturer s recommendations. The on the tilt and gravity recordings, it has been neglected in the tilt zero sensor position is obtained by hardware tuning using following computations. footscrews. The tilt calibration factors in the x and y directions On Scintrex CG-3/3M devices, the tilt parameter is usually are then experimentally defined by comparing the gravity used to apply real-time corrections to the gravity measure- readings obtained for extreme tilt values of about 150 arcsec. ments. Any atmospheric-pressure effect on the tilt meters will The correction factor for each tilt correction constant is

13 Continuous gravity recording 481 Figure 10. Variations of the tilts along the X and Y axes of the three side-by-side gravimeters (Bondy, from 01/16/97 to 02/20/97). The jump at day 34 corresponds to a manual resetting of the tilts. Instruments and were equipped with high-resolution tiltmeters (0.1 arcsec). A correlation with atmospheric-pressure variations can be observed at short wavelengths (see Fig. 7). expressed as K=S (R 0 R 1 ), (3) X2 1 The results presented above show that the influence of the atmospheric pressure on the internal temperature and tilt parameters could induce non-negligible gravimetric variations at short periods during real-time data processing. Nonetheless, these instrumental effects can be removed from the gravimet- ric recordings later, if local meteorological recordings made simultaneously are available. where R and R are the respective gravity readings in mgal 0 1 taken at tilt values X =0 and X =±150 arcsec. 0 1 The absence of correlation between any of the gravimetric and inclinometric recordings (Figs 5, 9 and 10) ensures, on one hand, that the tilt corrections have been correctly computed in the real-time processing and, on the other hand, that the tilt zero adjustment and tilt sensitivity have remained stable during the whole period of data acquisition. This latter point can be clearly checked on day 34, where the resetting of all tilt sensors to zero is not associated with a jump in the corresponding gravity recordings (Figs 5 and 10). Fig. 10 also shows a higher resolution of tilt recordings for meters and The standard 1 arcsec resolution has been improved to 0.1 arcsec by using new software to check the actual accuracy of the tilt sensors. On the basis of these responses, it can be verified that the instrumental noise is lower than 1 arcsec. One of the reasons for the apparent accuracy of the tilt sensors comes from the fact that the bubble levels are co-located with the gravity sensor, in the thermo- statically controlled enclosure, and are therefore isolated from any temperature variation. Results from additional tests performed on the tilt resolution are given later in this paper. 4.3 Residual gravity responses L ong-term temperature correction Over long periods, a correlation appears between the residual gravity signals and the internal temperature. It may indicate the persistence of an instrumental effect (Figs 5, 8 and 9). In order to reduce this effect, the transfer functions between these two parameters were computed in the frequency domain. The resulting admittance function is given as follows, for a given

14 482 S. Bonvalot, M. Diament and G. Gabalda Figure 11. Influence of the atmospheric pressure on the gravimeter tilt-meter responses (Bondy, from 01/16/97 to 02/20/97). Thick lines show uncorrected tilts and thin lines show tilts corrected using the coefficient determined for each meter by a linear regression between the two parameters (see text and Table 1). frequency f: G( f ) T *( f ) Z( f )= For each instrument, a mean value of the admittance was computed by averaging the values obtained for periods corresponding to coherence values greater than 0.5 (periods greater T( f ) T *( f ), (4) than 5 or 6 days). The results are shown in Table 1. These where G( f ) and T ( f ) are the energy spectra of the gravimetric values reflect the long-term correlation between the thermal and thermal signals, respectively, and T *( f ) is the complex variations and the gravimetric variations. This correlation conjugate of T ( f ). cannot be taken into account in the correction factor TEMPCO The energy spectra were computed by a Fourier transform which is determined experimentally over very short periods of after removal of the linear trend. High frequencies were filtered time (a few hours). Gravity recordings corrected for longby applying a smoothing polynomial filter. The parameters period temperature variations using admittance values deter- applied for this filtering were determined by computing the mined in this study are reported in Fig. 13. The effect of the maximum of the correlation corresponding to the largest removal of this long wavelength is clearly seen in the signals values of coherence between a pair of signals. For each and significantly improves the gravity recordings. This obser- meter, the coherence, as well as the corresponding admittance vation proves that a more accurate signal can be obtained by and phase functions, is shown in Fig. 12. Coherence values post-processing the data while looking for the correlation with greater than 0.5 are observed for the lower frequencies ( below the thermal signals recorded simultaneously. 4 or 5 cycles day 1), before a sharp drop in the coherence. Within this same low-frequency band, constant values of the admittance and null values of the phase are obtained for each meter, indicating a very good correlation between the gravity and temperature signals. Correction for the direct evect of the atmospheric pressure So far, the gravimetric signals have been corrected for the indirect effects of the atmospheric pressure (through the temperature

15 Continuous gravity recording 483 Figure 12. Transfer functions (coherence, admittance and phase) between the gravity and the thermal signals computed in the frequency domain for each Scintrex CG-3M gravimeter between 01/16/97 and 02/20/97. A pre-processing was applied to the signals in order to remove the linear trend and to filter high frequencies (see text). This figure shows that the coherence is greater than 0.5 for all meters for time periods greater than 5 6 days (frequency lower than 5). The admittance is flat in this domain and the phase is close to zero. Therefore, an average coefficient (in mgal mk 1) can be obtained and later used in post-processing (see Table 1). This coefficient differs from the short term one given by the manufacturer. Frequency unit: cycles day 1; period (day)=k/ f, with k= or the tilt), but the direct influence of the atmospheric pressure has not been taken into account. The influence of the atmospheric pressure on the gravity field, as pointed out by several authors (Warburton & Goodkind 1977; Merriam 1992), is induced by the weight of the air column and can vary between 0.2 and 0.4 mgal hpa 1. The related correction can be computed using a standard model or a value defined from simultaneous recordings of the pressure and of the gravity field. A standard value of mgal hpa 1 (Merriam 1992) was applied to correct the gravity recordings for the atmospheric-pressure variations observed near the surveyed area (Fig. 14). Here, the maximal amplitude of the direct effect of the atmospheric pressure is less than 15 mgal over the period of time considered. Taking this influence into account improves the quality of the observed gravity signals, particularly at the short wavelengths. 4.4 Differential signal computation The residual gravity signals obtained after applying the various corrections show a noise of the order of 0.01 mgal as well as comparable amplitude variations for the short and long wavelengths. These variations can be related to effects not taken into account or not properly corrected during data processing (insufficiently accurate Earth tide or atmospheric-pressure corrections, etc.). Given that these effects might be considered

16 484 S. Bonvalot, M. Diament and G. Gabalda Figure 13. Gravity responses corrected for the instrumental thermal effects using the long-term coefficient (see Fig. 12 and Table 1). Comparison with Fig. 5 shows that the residual is flatter after the removal of this long term thermal effect. as constant over an area of a few square kilometres, they can Differences lower than ±15 mgal are observed in the gravity be easily eliminated by applying a difference between residual differential measurements over more than a month of gravity signals recorded on several instruments not too far recording. Better results are obtained for the difference between apart. gravimeters and (Fig. 15b) because their In order to analyse these differential signals, the instruments instrument responses are more homogeneous. If the highfrequency should be precisely calibrated. For this purpose, several noise in the gravimetric signals is filtered first methods can be used, for instance the measurement of a (Fig. 14), the differences observed with respect to the reference calibration line or the computation of the admittance between gravimeter are of the order of ±5 and ±10 mgal for the theoretical and observed tide signals. The latter method does gravimeters (Fig. 15b) and (Fig. 15a), not require the movement of the instruments and seems to be respectively. A similar study performed on shorter time-series particularly well adapted to the analysis of signals recorded allowed to verify that the deviations could be reduced to ±2 on permanent gravimeter networks. An example is given by to ±3 mgal over a few days. As previously noticed, the Jousset et al. (in press) for the processing of the time-series standard errors responses (deviation <±3 mgal) and the tilt recorded on the Merapi volcano. Here we calibrated the response (deviation <±3 arcsec) are remarkably homo- various instruments on a calibration line covering the measure- geneous. The observed temperature differences remain below ment range of the continuous recordings. Calibration deviations 0.05 mk. observed between different instruments were smaller Therefore, we conclude that Scintrex CG-3M gravimeters in than 5 mgal. The details and results of these computations are adifferential mode can allow the detection of gravity field time discussed later in this paper. variations with an accuracy up to 5 15 mgal over time periods Using residual signals properly corrected for the instrumental of several weeks. However, this accuracy, obtained under effects and for the calibration discrepancies determined as particularly favourable conditions, is slightly better than the described above, differences in the recorded parameters (gravity actual resolution classically obtained in field conditions (Rymer field, standard error, temperature, tilts) between instruments 1989; Torge 1989). This study was performed with the standard were computed over the same time period. Figs 15(a) and (b) seismic-noise filter. We might expect more accurate results show the differences observed for gravimeters and using the filter proposed in 1997 by Scintrex for the new , respectively, using instrument as a reference. CG-3/3M software. According to the technical specifications,

17 Continuous gravity recording 485 Figure 14. Gravity responses corrected for the atmospheric pressure using a standard value of mgal hpa 1. The dashed lines correspond to the signals filtered for high-frequency variations by applying a polynomial smoothing. this new filter could decrease the noise of the gravity recording by a factor of at least 5 times during periods of high seismicity. 5 GRAVIMETER CALIBRATION 5.1 Set-up The calibration factors of the gravimeters used in this study were checked in the same measurement range using the calibration line of the Bureau International des Poids et Mesures (BIPM), Sèvres, France. This line was established in 1994 during the fourth intercomparison of absolute and relative gravimeters (Becker et al. 1995) using fourteen LaCoste & Romberg relative gravimeters (D and G models). It includes six stations and covers a measurement range of about 8 mgal. Measurements were made on 1997 February 20 taking as a reference one of the BIPM absolute measurement stations. All measurement sites were used at least twice with four Scintrex CG-3M gravimeters. The measurements were corrected for the Earth tide, taking into account precise amplitude and phase coefficients computed earlier for this site (Ducarme, personal communication). using Scintrex devices and the reference values (mean values obtained in 1994 with fourteen LaCoste & Romberg gravimeters). Fig. 16(a) shows the residuals between these two sets of values as observed at each station, as well as the calibration factors computed for the different meters. The largest difference between the values obtained on Scintrex instruments and the reference is about mgal. It can be noticed that the results obtained with the different meters are very consistent: the calibration factors are lower than 1 and the residues observed at each station have the same sign. This observation, already made in the past (Jousset et al. 1995), could be related to a slightly different sensitivity of the Scintrex instruments from that of the LaCoste & Romberg gravimeters (magnetic effect). In order to verify the homogeneity of the Scintrex CG-3M instruments, relative calibration coefficients were also computed using one of the instruments ( ) as a reference. The residual deviations observed between the recalibrated devices and a reference defined as a mean over the four instruments are shown in Fig. 16( b). In this study, for the three instruments used in the computation of the differential signals, the calibration deviations remained below or equal to 5 mgal. 5.2 Calibration results Calibration factors were computed by applying a linear regression between the observed values obtained at each station 6 DISCUSSION The analysis of continuous data recorded on three Scintrex CG-3M instruments allows us

18 486 S. Bonvalot, M. Diament and G. Gabalda Figure 15. Differential instrumental responses for the gravimeters. (a) Responses corresponding to the differences between the signals acquired on gravimeters and at Bondy between 01/16/97 and 02/20/97: curve a, difference between the gravity signals; curve b, difference between the filtered gravity signals (see Fig. 14); curve c, difference between the measurement errors; curve d, difference between the tilt signals; curve e, difference between the thermal signals. (b) Responses corresponding to the difference between signals acquired on gravimeters and at Bondy between 01/16/97 and 02/20/97; curves a e as in part (a). (1) to define some of the instrument characteristics; this well-known direct influence, these variations can affect (2) to evaluate the accuracy of the measurements when the significantly the gravity signal through the automatic correcinstruments are used in a differential mode. Several aspects specific to the instrument characteristics and accuracy can be discussed qualitatively and quantitatively. 6.1 Instrumental artefacts It has been noted that some instruments could show a strong sensitivity to atmospheric-pressure variations. In addition to tions of the temperature and the tilt. Nonetheless, as we have shown in this study, this dependence can be easily controlled and corrected during the post-processing by looking for a possible correlation with simultaneous weather recordings. In the same way, a long-term correlation exists between the gravimetric signal and the internal temperature of the device. The thermal correction applied in real time by the Scintrex CG-3/3M gravimeter cannot take into account such effects revealed over longer periods of time; only a post-processing of the data series allows one to correct these effects.

19 Continuous gravity recording 487 Figure 15. (Continued.) It is necessary to determine the instrumental effects for each instrument prior to any study based on continuous microgravity 6.2 Accuracy of the instrumental responses recordings that is intended to investigate temporal gravity The accuracy of the measurements made with the Scintrex variations with a geodynamic origin. The importance of simultaneously CG-3M gravimeters has been estimated with four instruments acquiring the largest possible number of parameters in two utilization modes, as follows. affecting the quality of the instrument responses (atmospheric pressure, temperature, tilt, etc.) has already been pointed out (1) In-field discrete measurements: during the calibration by Goodkind (1986). This point demonstrates the added value using the reference stations of the BIPM, an accuracy of about of the Scintrex CG-3/3M, which systematically records several 5 mgal was obtained over a measurement range of 8 mgal. This of these parameters (temperature and tilt). Given the importance result has been confirmed by other calibration surveys carried of the atmospheric pressure in this type of study, we out over wider ranges of measurement in the frame of micro- recommend that sensors recording the surrounding pressure gravity studies applied to volcanology (Budetta & Carbone be integrated in the next generation of microgravimeters. The 1997). recording of this parameter could then be used to correct (2) In continuous recording mode: this study shows that gravity series in real time or at a later stage in the processing. several instruments can reveal time variations of small amplitude

20 488 S. Bonvalot, M. Diament and G. Gabalda Figure 16. Calibration results for four gravimeters on the calibration baseline of the Bureau International des Poids et Mesures (02/20/97). The coefficients k and sd are the computed gravimetric correction factor and the standard deviation in mgal of the adjustment respectively. The gravity values at the stations numbered are referenced to an absolute gravity base station. (a) Deviations obtained with respect to a reference defined with 14 LaCoste & Romberg gravimeters in 1994 (reference defined during the fourth international comparison of absolute gravimeters). ( b) Deviations obtained with respect to a reference defined as an average over four Scintrex CG-3M gravimeters (instrument used as a reference). (lower than 5 mgal or 15 mgal for time periods or a few days to a few weeks, respectively). This result is of particular 6.3 Long-term evolution of the instrumental drift interest in volcano monitoring, where microgravity variations It has been demonstrated that the instrumental drift of Scintrex can be observed over very short time periods before and after CG-3/3M gravimeters over time periods of several weeks can phases of activity (Rymer & Brown 1989; Rymer 1994, 1995; be modelled with a quadratic function. This implies that the Jousset et al. in press). A more general usage of microgravimeter wavelength of the gravity signal being recorded should not be networks should improve the detection capabilities for comparable to that of the quadratic signal characterizing the volcanic-eruption precursor signals and their understanding. instrumental drift. If this drift is stable enough, it should be

21 Continuous gravity recording 489 possible to study long-term gravity variations. In order to strated by the discrete drift values computed from the corresponding quantify better the changes in the instrumental drift of the continuous recordings (Fig. 17b). These values, Scintrex CG-3/3M gravimeters over longer periods of time, defined over periods of several days, show a sharp variation the response of two instruments was analysed over time since between the 200th and 300th days of use, followed by a slower their date of acquisition. Gravity field values, recorded at the and more regular decrease of the drift rate of both instruments. same reference site in Bondy by gravimeters and After several years, the values are about 0.25 mgal day 1 and are shown in Fig. 17(a). It can be noted that these 0.35 mgal day 1 for instruments and , respectively. instruments have drifted by about 1070 mgal over 2400 days These values are equivalent to those determined in the and 250 mgal over 700 days, respectively. The corresponding framework of this study (Fig. 4 and Table 1). The two instruments daily averages are 0.45 mgal and 0.36 mgal, respectively. This have different behaviours: they have drifted in opposite drift did not occur in a linear manner over time, as demon- directions since the first phase of utilization (inversion of the Figure 17. Long-term instrumental drift for the Scintrex CG-3M gravimeters. (a) Gravity field values measured at the reference site in Bondy with instruments and since their first use. The jump seen around day 1600 follows the opening of the sensor during a control by the manufacturer. (b) Corresponding instrumental drift values computed over continuous recording periods greater than or equal to three days.

22 490 S. Bonvalot, M. Diament and G. Gabalda drift for ). These observations confirm that sharp variations in the instrumental drift can occur during the first year of use, as mentioned by the manufacturer (Scintrex 1995). In the long term, the evolution of the instrumental drift is more regular. If a regular instrument calibration is performed, gravimetric variations of longer periods could be also studied through continuous recordings. Figure 18. Schematic view of the tilt calibration table. The circled cross indicates the location of the mercury step-bearing. Axes x and 1 x indicate the position of the gravimeter X axis during tests 1 and 2 2 shown in Fig Tiltmeter resolution It was proved above that the accuracy of the tilt sensor was greater than the proposed standard resolution (1 arcsec). To check the sensitivity of the sensors to a small tilt variation, one gravimeter was placed on a calibration table normally used for the calibration of quartz tilt sensors (P. A. Blum, personal communication). The measurement system, shown in Figure 19. Results of an experimental test on the tilt response of the gravimeter (sampling rate=1 reading per 90 sec). Curve a, tilts induced in the X axis (test 1) and the Y axis (test 2) of the gravimeter (see Fig. 18). In both tests, the calibration table was tilted alternately from position 1 (tilt=0) to position 2 (tilt=4 mrad). Curve b, tilt response of the gravity meter along the X axis. Curve c, tilt response of the gravity meter along the Y axis. Curve d, standard error of the gravity measurement series. Curve e, internal temperature of the gravimeter.